CN114842447A - Convolutional neural network-based parking space rapid identification method - Google Patents
Convolutional neural network-based parking space rapid identification method Download PDFInfo
- Publication number
- CN114842447A CN114842447A CN202210486042.5A CN202210486042A CN114842447A CN 114842447 A CN114842447 A CN 114842447A CN 202210486042 A CN202210486042 A CN 202210486042A CN 114842447 A CN114842447 A CN 114842447A
- Authority
- CN
- China
- Prior art keywords
- parking space
- network
- convolutional neural
- neural network
- angular points
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/586—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of parking space
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention provides a parking space rapid identification method based on a convolutional neural network, which comprises the following steps: s1, acquiring a parking space image through a wide-angle camera, and converting the parking space image into a top view through inverse perspective transformation; s2, constructing a light parking space detection network through Ghost convolution, deep separable convolution and the like; s3, positions of the parking space angular points, the entrance lines and the separation lines are predicted simultaneously through a convolutional neural network; and S4, pairing the parking space angular points according to the network prediction result, and reasoning out the complete parking space. By applying the technical scheme, the lightweight deep learning convolutional neural network is constructed to predict the positions of the parking space angular points, the entrance lines and the separation lines, the parking space angular points are matched by utilizing the position relation of the entrance lines and the parking space angular points, the parking space positions are determined, and then the complete parking space is deduced.
Description
Technical Field
The invention relates to the technical field of automatic driving of automobiles, in particular to a parking space quick identification method based on a convolutional neural network.
Background
In recent years, with the rapid development of the automobile industry in China and the continuous improvement of the living standard of people, the number of automobiles in cities is continuously increased, and the problem of difficult parking is more and more prominent. Thanks to the rapid development of computer technology, the automatic driving technology of vehicles has also been greatly improved, wherein the automatic parking function solves the problem of difficult parking to a certain extent and is popular with users. In the automatic parking technology, what needs to be solved first is the detection and positioning of parking spaces. At present, various methods are available at home and abroad to realize detection of parking spaces, including ultrasonic radar, laser range finders, ground induction coils, laser radar, computer vision and the like. Except for computer vision, other methods have single function, difficult maintenance, difficult installation and higher cost.
For a visual parking space detection method, the former people have many researches: zhang et al put forward a deep learning-based parking space detection method deep PS in Vision-based park-slot detection, A DCNN-based assessment and a large-scale coordinate dataset, firstly detecting parking space corner points by using YOLOv2, and then obtaining parking space types and parking space directions through local image classification network and template matching. Jiang et al in the Detection of park Slots Based on Mask R-CNN, using semantic segmentation to detect the Parking space, using the Mask R-CNN to detect the corner points of the Parking space, and generating a Mask, and then using post-processing methods such as line segment Detection to extract and combine the vehicle position lines to infer the Parking space. Li et al, in "Vacand parking stall detection in the around view image based on parking stall learning", apply YOLOv3 to detect the head and corner of the parking stall, and then use the prior geometric information to deduce the parking stall, but this method needs a complicated rule-based scheme to deduce the parking stall direction, and the detection process is more complicated to calculate, and is difficult to use in engineering practice.
Disclosure of Invention
In view of the above, the present invention aims to provide a method for quickly identifying a parking space based on a convolutional neural network, which constructs a lightweight deep learning convolutional neural network to predict positions of a parking space angular point, an entrance line and a separation line, pairs the parking space angular points according to a position relationship between the entrance line and the parking space angular points, determines the parking space positions, and further deduces a complete parking space.
In order to achieve the purpose, the invention adopts the following technical scheme: a parking space rapid identification method based on a convolutional neural network comprises the following steps:
s1, acquiring a parking space image through a wide-angle camera, and converting the parking space image into a top view through inverse perspective transformation;
s2, constructing a lightweight parking space detection network through Ghost convolution, deep separable convolution and the like;
s3, positions of the parking space angular points, the entrance lines and the separation lines are predicted simultaneously through a convolutional neural network;
and S4, pairing the parking space angular points according to the network prediction result, and reasoning out the complete parking space.
In a preferred embodiment: in step S1, the image obtained by the wide-angle camera is first subjected to distortion removal processing, then a parking space top view is obtained through inverse perspective transformation, the length-width ratio and the size of the top view are designed according to the physical information of the real world marker, the mapping relationship from the pixel coordinate to the real world coordinate is realized, and overhead parking space images in various environments are collected.
In a preferred embodiment: in step S2, a Ghost convolutional stack is used to form a Ghost bottleeck module, and an ECA attention mechanism is introduced to form an ECA-Ghost bottleeck module, so that the network detection effect is improved without increasing the number of network parameters and the amount of computation; then an ECA-GhostBottleneck module and the residual edge are combined to form an ECA-GhostC3 module for parking space feature extraction; while the embedding depth separable convolution is used to scale the feature map.
In a preferred embodiment: and setting two detection output layers according to the size of the detection target.
In a preferred embodiment: in step S3, adding a separation line direction regression branch, and directly outputting the parking space direction through a network; and making a network training data set according to the predicted target; after the network training is completed, the number m of the feature map channels output by the network can be represented by the following formula:
m＝(5+n c +o)n a
in the formula, n c 2 is the number of detection categories; n is a 3, the number of anchor frames of the Grid Cell; the number m of channels in the network output characteristic diagram is (5+2+2) × 3 is 27.
In a preferred embodiment: in the aspects of positioning the parking space corner points and the entrance line target frames, alpha-IoU is used as a loss function; using cross entropy as a loss function in terms of target classification and confidence; use of Smooth L in the branch of the separation line azimuthal regression 1 As a loss function;
wherein O is the deviation between the predicted value and the true value of the network, Loss Orientation Is a loss function of a separation line azimuth regression branch of the network, K multiplied by K represents the number of grids into which the input parking space image is divided, i represents the ith grid,
in order to predict the value of the azimuth,
and
the tag is the true value and activated using the sigmoid function.
In a preferred embodiment: in step S4, the corner points are paired according to the positional relationship between the parking space corner points and the entrance lines in the prediction result, and the parking space position is determined according to the position of the separation line connected to the paired corner points.
In a preferred embodiment: after the parking space angular points are matched, the parking space type is determined according to the distance between the two angular points, and then the complete parking space is deduced according to the depths of the prior parking spaces of different types.
Compared with the prior art, the invention has the following beneficial effects:
1. the method for quickly identifying the parking space based on the convolutional neural network can automatically provide accurate parking space position information for a parking system.
2. A lightweight parking space detection network is constructed through Ghost convolution, deep separable convolution and the like, compared with a mainstream target detection network, the calculation amount and the parameter amount are greatly reduced, and the dependence on hardware calculation resources is reduced.
3. And a parking space separation line azimuth regression branch is added in the network, so that a complex post-processing stage is avoided, and complete parking space information can be directly deduced according to a network prediction result.
Drawings
FIG. 1 is a flow chart of parking space identification according to a preferred embodiment of the present invention;
FIG. 2 is a schematic view of a typical parking space type according to the preferred embodiment of the present invention;
FIG. 3 is a diagram of a parking space recognition network according to a preferred embodiment of the present invention;
FIG. 4 is a block diagram of a parking space recognition network according to a preferred embodiment of the present invention;
FIG. 5 is a diagram of the predicted elements of the parking space recognition network in accordance with the preferred embodiment of the present invention;
fig. 6 is a schematic diagram of a complete parking space inference in the preferred embodiment of the present invention.
Detailed Description
The invention is further explained below with reference to the drawings and the embodiments.
It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application; as used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
A method for quickly identifying a parking space based on a convolutional neural network refers to FIGS. 1 to 6, and includes the following steps:
s1, acquiring a parking space image through a wide-angle camera, and converting the parking space image into a top view through inverse perspective transformation;
s2, constructing a lightweight parking space detection network through Ghost convolution, deep separable convolution and the like;
s3, positions of the parking space angular points, the entrance lines and the separation lines are predicted simultaneously through a convolutional neural network;
and S4, pairing the parking space angular points according to the network prediction result, and reasoning out the complete parking space.
Specifically, in step S1, the image obtained by the wide-angle camera is first subjected to distortion removal processing, then a parking space top view is obtained through inverse perspective transformation, the length-width ratio and the size of the top view are designed according to the physical information of the real world marker, the mapping relationship from the pixel coordinates to the real world coordinates is realized, and overhead parking space images in various environments are collected.
In step S2, a Ghost convolutional stack is used to form a Ghost bowtleneck module, and an ECA attention mechanism is introduced into the Ghost bowtleneck module to form an ECA-Ghost bowtleneck module, so as to improve the network detection effect without increasing the number of network parameters and the amount of computation; then an ECA-GhostBottleneck module and the residual edge are combined to form an ECA-GhostC3 module for parking space feature extraction; while the embedding depth separable convolution is used to scale the feature map. Finally, a lightweight parking space detection network is constructed as shown in fig. 3; the Ghost convolution, ECA-Ghost Bottleneck and ECA-Ghost C3 are formed as shown in FIG. 4, the network parameter is 2.25 × 106, the calculated amount is 4.5GFLOPs, and the calculation amount is far lower than that of a common target detection network.
And setting two detection output layers according to the size of the detection target.
In step S3, in addition to the target detection branch for detecting the parking space corner point and the entrance line, a separation line orientation regression branch is added, and the parking space orientation is directly output through the network; and making a network training data set according to the predicted target; after the network training is completed, the elements of the network prediction output are shown in fig. 5, wherein Cx and Cy represent the center of the prediction box, W, H represents the width and height of the prediction box, P0 represents the confidence of the object, P1 and P2 represent the prediction class probability, respectively, and Ox and Oy represent the predicted orientation information. The number m of signature channels output by the network can be represented by the following formula:
m＝(5+n c +o)n a
in the formula, n c 2 is the number of detection categories; n is a 3, the number of anchor frames of the Grid Cell; the number m of channels in the network output characteristic diagram is (5+2+2) × 3 is 27.
In the aspects of positioning the corner points of the parking spaces and the target frames of the entrance lines, alpha-IoU is used as a loss function to achieve higher positioning accuracy; using cross entropy as a loss function in terms of target classification and confidence; use of Smooth L in the branch of the separation line azimuthal regression 1 As a loss function;
wherein O is the deviation between the predicted value and the true value of the network, Loss Orientation Is a loss function of a separation line azimuth regression branch of the network, K multiplied by K represents the number of grids into which the input parking space image is divided, i represents the ith grid,
in order to predict the value of the azimuth,
and
the tag is the true value and activated using the sigmoid function.
In step S4, the corner points are paired according to the positional relationship between the parking space corner points and the entrance lines in the prediction result, and the parking space position is determined according to the position of the separation line connected to the paired corner points.
After the parking space angular points are matched, the parking space types are determined according to the distance between the two angular points, and then the complete parking spaces are deduced according to the depths of the prior different parking spaces, as shown in fig. 6, wherein m1 and m2 are successfully paired parking space angular points, m3 and m4 are the other two invisible angular points of the parking spaces.
Claims (8)
1. A parking space rapid identification method based on a convolutional neural network is characterized by comprising the following steps:
s1, acquiring a parking space image through a wide-angle camera, and converting the parking space image into a top view through inverse perspective transformation;
s2, constructing a lightweight parking space detection network through Ghost convolution, deep separable convolution and the like;
s3, positions of the parking space angular points, the entrance lines and the separation lines are predicted simultaneously through a convolutional neural network;
and S4, pairing the parking space angular points according to the network prediction result, and reasoning out the complete parking space.
2. The method for quickly identifying the parking space based on the convolutional neural network as claimed in claim 1, wherein the method comprises the following steps: in step S1, the image obtained by the wide-angle camera is first subjected to distortion removal processing, then a parking space top view is obtained through inverse perspective transformation, the length-width ratio and the size of the top view are designed according to the physical information of the real world marker, the mapping relationship from the pixel coordinate to the real world coordinate is realized, and overhead parking space images in various environments are collected.
3. The method for quickly identifying the parking space based on the convolutional neural network as claimed in claim 1, wherein the method comprises the following steps: in step S2, a Ghost convolutional stack is used to form a Ghost bottleeck module, and an ECA attention mechanism is introduced to form an ECA-Ghost bottleeck module, so that the network detection effect is improved without increasing the number of network parameters and the amount of computation; then an ECA-GhostBottleneck module and the residual edge are combined to form an ECA-GhostC3 module for parking space feature extraction; while the embedding depth separable convolution is used to scale the feature map.
4. The method for quickly identifying the parking space based on the convolutional neural network as claimed in claim 3, wherein the method comprises the following steps: and setting two detection output layers according to the size of the detection target.
5. The method for quickly identifying the parking space based on the convolutional neural network as claimed in claim 1, wherein the method comprises the following steps: in step S3, adding a separation line direction regression branch, and directly outputting the parking space direction through a network; and making a network training data set according to the predicted target; after the network training is completed, the number m of characteristic diagram channels output by the network is represented by the following formula:
m＝(5+n c +o)n a
in the formula, n c 2 is the number of detection categories; n is a 3, the number of anchor frames of the Grid Cell; the number m of channels in the network output characteristic diagram is (5+2+2) × 3 is 27.
6. The method for quickly identifying the parking space based on the convolutional neural network as claimed in claim 5, wherein the method comprises the following steps: in the aspects of positioning the parking space corner points and the entrance line target frames, alpha-IoU is used as a loss function; using cross entropy as a loss function in terms of target classification and confidence; use of Smooth L in the branch of the separation line azimuthal regression 1 As a loss function;
wherein O is the deviation between the predicted value and the true value of the network, Loss Orientation Is a loss function of a separation line azimuth regression branch of the network, K multiplied by K represents the number of grids into which the input parking space image is divided, i represents the ith grid,
in order to predict the value of the azimuth,
and
the tag is the true value and activated using the sigmoid function.
7. The method for quickly identifying the parking space based on the convolutional neural network as claimed in claim 1, wherein the method comprises the following steps: in step S4, the corner points are paired according to the positional relationship between the parking space corner points and the entrance lines in the prediction result, and the parking space position is determined according to the position of the separation line connected to the paired corner points.
8. The method according to claim 7, characterized in that: after the parking space angular points are matched, the parking space type is determined according to the distance between the two angular points, and then the complete parking space is deduced according to the depths of the prior parking spaces of different types.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210486042.5A CN114842447A (en) | 2022-05-06 | 2022-05-06 | Convolutional neural network-based parking space rapid identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210486042.5A CN114842447A (en) | 2022-05-06 | 2022-05-06 | Convolutional neural network-based parking space rapid identification method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114842447A true CN114842447A (en) | 2022-08-02 |
Family
ID=82568122
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210486042.5A Pending CN114842447A (en) | 2022-05-06 | 2022-05-06 | Convolutional neural network-based parking space rapid identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114842447A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117152716A (en) * | 2023-09-13 | 2023-12-01 | 哈尔滨理工大学 | Parking space detection method considering direction entry line and auxiliary mark point |
| WO2024032856A1 (en) * | 2022-08-12 | 2024-02-15 | Continental Autonomous Mobility Germany GmbH | Method for determining a parking space and a target position for a vehicle in the parking space |
-
2022
- 2022-05-06 CN CN202210486042.5A patent/CN114842447A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2024032856A1 (en) * | 2022-08-12 | 2024-02-15 | Continental Autonomous Mobility Germany GmbH | Method for determining a parking space and a target position for a vehicle in the parking space |
| CN117152716A (en) * | 2023-09-13 | 2023-12-01 | 哈尔滨理工大学 | Parking space detection method considering direction entry line and auxiliary mark point |
| CN117152716B (en) * | 2023-09-13 | 2024-04-02 | 哈尔滨理工大学 | Parking space detection method considering direction entry line and auxiliary mark point |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | Springrobot: A prototype autonomous vehicle and its algorithms for lane detection | |
| CN114842447A (en) | Convolutional neural network-based parking space rapid identification method | |
| CN107103277B (en) | Gait recognition method based on depth camera and 3D convolutional neural network | |
| CN113673420B (en) | Target detection method and system based on global feature perception | |
| CN109948707B (en) | Model training method, device, terminal and storage medium | |
| CN110766723B (en) | Unmanned aerial vehicle target tracking method and system based on color histogram similarity | |
| CN112907602B (en) | Three-dimensional scene point cloud segmentation method based on improved K-nearest neighbor algorithm | |
| CN109934857B (en) | Loop detection method based on convolutional neural network and ORB characteristics | |
| CN112232139B (en) | Obstacle avoidance method based on combination of Yolo v4 and Tof algorithm | |
| CN113095152A (en) | Lane line detection method and system based on regression | |
| CN112396036A (en) | Method for re-identifying blocked pedestrians by combining space transformation network and multi-scale feature extraction | |
| CN113406659A (en) | Mobile robot position re-identification method based on laser radar information | |
| CN115497122A (en) | Method, device and equipment for re-identifying blocked pedestrian and computer-storable medium | |
| CN114926796A (en) | Bend detection method based on novel mixed attention module | |
| CN112084895A (en) | Pedestrian re-identification method based on deep learning | |
| CN111354016A (en) | Unmanned aerial vehicle ship tracking method and system based on deep learning and difference value hashing | |
| CN112528994A (en) | Free-angle license plate detection method, license plate identification method and identification system | |
| WO2015027649A1 (en) | Vehicle detection method using multi-scale model | |
| Zhou et al. | Obstacle detection for unmanned surface vehicles by fusion refinement network | |
| CN111860084B (en) | Image feature matching and positioning method and device and positioning system | |
| CN114155489A (en) | Multi-device cooperative unmanned aerial vehicle flyer detection method, device and storage medium | |
| CN113486723A (en) | Efficient vehicle re-identification method and device | |
| Liu et al. | Notice of retraction: Sharp curve lane detection for autonomous driving | |
| CN113836251A (en) | Cognitive map construction method, device, equipment and medium | |
| CN113610885B (en) | Semi-supervised target video segmentation method and system using difference contrast learning network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |